A significant problem with existing automated techniques for removing dust, scratches and other defects from an image is that such techniques often fail to adequately distinguish actual image features, such as line segments and various textures, from true defects, such as artifacts that arise when scanning a photograph. In addition, such existing techniques often are very poor at isolating low-contrast defects, especially where such low-contrast defects are located in darker regions. Due to the inherent limitations of such techniques, one often is faced with a choice between frequently missing such defects or frequently misclassifying image features as defects.